One of the few accepted treatments for severe non-fluent aphasia is Melodic Intonation Therapy (MIT). Inspired by the common clinical observation that patients can actually sing the lyrics of a song better than they can speak the same words, MIT emphasizes the prosody of speech through the use of slow, pitched vocalization (singing), and has been shown to lead to significant improvements in propositional speech beyond the actual treatment period. It has been hypothesized that this effect is due to the gradual recruitment of right- hemispheric language regions for normal speech production, and this is further supported by our own functional magnetic resonance imaging (fMRI) pilot data. Although the MIT-induced treatment effect has been shown in several small case series, it is not clear whether the effect is due to the intensity of the treatment or to the unique, components of MIT that are not found in other, non-intonation-based interventions. Thus, our overall aim is to test our hypothesis that MIT's rehabilitative effect is achieved by using its melodic and rhythmic elements to engage and/or unmask the predominantly right-hemispheric brain regions capable of supporting expressive language function. In order to test this hypothesis, we have developed an experimental design that includes the randomization of chronic stroke patients with persistent, moderate to severe non-fluent aphasia into three parallel groups receiving 1) 75 sessions of Melodic Intonation Therapy (approximately 8 weeks), 2) 75 sessions of an equally intensive, alternative verbal treatment method developed for this study (Speech Repetition Therapy), or 3) an equal period of No Therapy. All patients will undergo two pre-therapy and two post-therapy behavioral assessments in addition to the pre- and post-therapy fMRI studies examining the neural correlates of overtly spoken and sung words and phrases. This design allows us to 1) examine the efficacy of MIT over No Therapy, 2) examine the effects of elements specific to MIT (e.g., melodic intonation and rhythmic tapping) by comparing it to a control intervention (SRT) that is similar in structure and intensity of treatment, 3) compare post-therapy effects with pre-therapy baseline variations, and 4) examine post-treatment maintenance effects. Our primary speech outcome measure will be the number of Correct Information Units (CIU)/min produced during spontaneous speech. Secondary outcome measures include correctly named items on standard picture naming tests, timed automatic speech, and linguistically-based measures of phrase and sentence analysis.